1. INTRODUCTION

Raman spectroscopy has proven to be a versatile tool for the identification of liquids, solids, and gases. The extension of this technology to handheld systems for the detection and identification of trace concentrations of gas-phase chemicals is important for applications ranging from breath analysis to environmental monitoring to chemical agent detection. This progress, however, is hindered by the extremely weak Raman signal generated from a dilute vapor over short interaction lengths. Benchtop gas-phase Raman systems have been developed for real-time gas analysis [1], and higher sensitivities have been achieved using resonant cavities [2] and multipass cells [3]. However, further miniaturization and sensitivity enhancements are required for Raman spectroscopy to become viable for handheld trace gas analysis.

Over the past two decades, a number of efforts have focused on increasing measurement sensitivity within small interaction volumes. For example, solid-phase extraction media have been used to trap and concentrate organic contaminants from water and air prior to analysis by Raman backscattering [4]. Due to non-zero sorption and desorption times, this signal enhancement approach results in a delay in the signal response to changes in trace gas concentration. In another example, the analyte vapor and the Raman pump light are confined in a hollow-core waveguide, yielding large increases in Raman efficiency for the light scattering that takes place within the waveguide [5,6]. Miniaturization efforts based on this approach are challenging because of the need to pressurize the air sample for injection into the waveguide and the need for a vacuum pump to remove it. This waveguide efficiency enhancement concept for spontaneous Raman scattering dates back to 1972 with the report of spectral intensification factors of $\sim 10$ per meter of fiber for hollow quartz fibers filled with a liquid analyte [7]. Miniaturized modern demonstrations of this concept can be found in the field of microfluidics [8,9].

In the work presented here, we leverage both waveguiding and analyte concentration signal enhancement mechanisms by coating a highly evanescent optical rib waveguide with a thin, transparent, hypersorbent polymer. This polymer coating, acting as the upper cladding of the waveguide structure, partitions and concentrates analytes of interest within the optical field of the propagating mode. Raman scattering generated in the cladding of a waveguide, also known as evanescent-field Raman scattering, was first shown to be efficiently coupled into guided modes for plastic-coated glass fibers in 1978 [10,11]. Recently, Dhakal et al. [12,13] have calculated and measured the waveguide efficiency enhancement for evanescent-field Raman scattering of isopropyl alcohol drop coated on a bare strip waveguide.

With our sorbent polymer-coated rib waveguides, we demonstrate the detection of Raman scattering from parts-per-billion (ppb) analyte concentrations using a waveguide length less than 1 cm. Because the sorption and desorption processes are passive, no mechanical pumps are required, and the thinness of the sorbent layer acts to minimize the time for the vapor to diffuse and equilibrate within the layer and produce the signal response. The ppb detection limit in such a short optical interaction length is made possible by a number of important features. First, the density of target analyte molecules is increased inside the hypersorbent polymer by as much as a factor of ${10}^8$ [14] compared to the ambient environment. Second, a significant percentage of the guided-mode power propagates within the hypersorbent polymer coating. Third, the waveguide is coated along its length so the pump light continuously generates Raman scattering as it propagates, a large fraction of which is collected back into the guided modes of the structure. Our calculations indicate a $42\times$ per cm enhancement due to waveguiding effects which, when coupled with the highest (${10}^8$) partition coefficient from among the analytes tested, yields an overall internal Raman signal enhancement of $4.2\times {10}^9/\mathrm{cm}$ compared to free-space micro-Raman scattering of the trace gas. Other analytes experience the same $42\times$ per cm waveguide enhancement but a reduced concentration enhancement due to reduced partitioning in the polymer.

Functionalized waveguides for enhanced Raman scattering could eventually lead to inexpensive photonic integrated circuit (PIC) architectures that include the source and detector on the same chip. In addition, multiple sorbent coatings may be incorporated onto the same chip to establish an array, enabling sensing of a wider range of chemicals, including hydrocarbons, toxic industrial chemicals, or chemical agents.

2. DEVICE FABRICATION AND FUNCTIONALIZATION

The chip-scale rib waveguides used in this work comprised a 175 nm-thick silicon nitride (${\mathrm{Si}}_3{\mathrm{N}}_4$) layer deposited by low-pressure chemical vapor deposition (LPCVD) over a 5 μm-thick thermal silicon oxide layer, all on a silicon wafer. LPCVD ${\mathrm{Si}}_3{\mathrm{N}}_4$ allows for low-loss light propagation at near-infrared wavelengths. After a thin-film resist was patterned on the ${\mathrm{Si}}_3{\mathrm{N}}_4$ surface using an electron beam in fixed-beam moving stage mode, the straight 2 μm-wide rib waveguides were formed by a 100 nm-deep reactive ion etch. The chip was then laser scribed and cleaved along a silicon crystal plane to produce smooth waveguide ends (facets) without the need for additional polishing. The 9.6 mm-long cleaved sample was then cleaned using oxygen plasma ashing and a piranha rinse. Fabry–Perot fringe analysis [15] shows optical losses of $\sim 2\mathrm{dB}/\mathrm{cm}$ at wavelengths between 980 and 1600 nm for the waveguide mode excited in this study.

Chemically activating the waveguides for trace gas detection was achieved by coating them with a custom-designed hypersorbent polymer, HCSFA2 [14], which serves as the upper cladding of the waveguide structure. This carbosilane polymer incorporates two hexafluoroisopropanol (HFIP) functional groups per repeat unit to facilitate hydrogen bonding with important gas phase analytes, such as phosphonate ester nerve agents or nitroaromatics [16]. The polymer provides a high density of HFIP hydrogen-bond (HB) acidic groups for reversible binding to complementary HB basic analyte molecules, resulting in partition coefficients that depend strongly on analyte basicity. HCSFA2 was deposited from a cyclohexanone solution onto the waveguides using a rastered microcapillary tip [17]. The coating thickness varied along the waveguide length in the 1–2 μm range. An optical microscope image of the coated waveguide is shown in Fig. 1(a). We modeled the coated waveguide mode properties using finite-element analysis (Comsol Multiphysics, 2D Electromagnetic Waves) with wavelength-dependent refractive indices. Figure 1(b) shows the horizontal component of the calculated electric field within the coated waveguide cross section for the ${\mathrm{TE}}_{00}$ mode at the pump wavelength with refractive indices equal to 1.76, 1.99, and 1.47 for ${\mathrm{SiO}}_2$, ${\mathrm{Si}}_3{\mathrm{N}}_4$, and HCSFA2, respectively. The model shows that approximately 25% of the modal power propagates within the polymer layer and would thus be capable of interacting with sorbed molecules.

 

Fig. 1. (a) Top view: optical microscope image of the coated waveguides. The set of five horizontal dark lines are 2 μm-wide rib waveguides, and the shaded area with interference fringes is the sorbant-coated region. More than 98% of the length of the sample was coated with sorbent for this work. (b) Cross-sectional diagram of the coated waveguide structure overlaid with a two-dimensional color map surface plot of the horizontal component of electric field calculated using a finite element mode solver. (c) The experimental setup for collecting the Raman signal. PMF, polarization-maintaining fiber; BPF, bandpass filter; RO, refractive objective; SO, Schwarzschild reflective objective; LPF, long-pass filter; OAP, off-axis parabolic mirror.

Download Full Size | PDF

3. OPTICAL MEASUREMENTS

A schematic diagram of the experimental setup is shown in Fig. 1(c). The coated waveguides were mounted in a flowcell—a sealed rectangular cuboid enclosure with glass windows on three sides and a tubing connector on each end that allowed for controlled gas flow perpendicular to the clear optical path through the system. The flowcell was purged continuously with nitrogen gas. To add a chemical vapor for analysis, a separate line of nitrogen gas was bubbled through the liquid chemical, and the resulting analyte-saturated nitrogen was combined with the continuously flowing nitrogen using a gas flow valve. Concentrations at the sample location were regulated using mass-flow controllers. The three analytes chosen for this proof-of-principle work were ethyl acetate (EA), methyl salicylate (MeS), and dimethyl sulfoxide (DMSO), listed in increasing order of partitioning into HCSFA2. These molecules provide different HB basicities deriving from the carboxylate and sulphoxide structures. As such, they serve as benign substitutes for the more toxic phosphonate esters, which exhibit strong HB basicity deriving from the phosphoryl oxygen.

The polarized output of a 1064 nm, single longitudinal mode, diode-pumped ${\mathrm{Nd}\text{:}\mathrm{YAG}/\mathrm{Nd}\text{:}\mathrm{YVO}}_4$ laser was first passed through a narrow optical bandpass filter to minimize the amount of light entering the waveguide at other frequencies. This pump light (43 mW) was then focused through the flow cell window and onto a waveguide facet using a $50\times$ long-working-distance refracting microscope objective. The polarization of the light was in the plane of the ${\mathrm{Si}}_3{\mathrm{N}}_4$ layer so as to excite the fundamental quasi-transverse electric mode, ${\mathrm{TE}}_{00}$. The pump light and forward-propagating Raman signal exited the back facet of the waveguide, passing through the opposite flow cell window and were collimated with a reflecting microscope objective. A reflecting objective was used to reduce chromatic dispersion in the light collection path. After passing through a long-pass edge filter used to block the remaining pump light ($\sim 0.3\mathrm{mW}$ typical), the signal was coupled into a 0.75 m spectrograph with a 300-groove/mm, 1.3 μm-blaze grating and a liquid nitrogen cooled, 1024 element InGaAs linear array detector. Spectra were collected for 100 s at each of five different grating positions. Post-collection processing included stitching the multiple grating position spectra together and performing a five-point central moving average.

Typical Raman scattering spectra, before and during exposure to 540 ppb DMSO, are shown in Fig. 2. The spectrum during exposure to DMSO is shifted upward in the main plot area for clarity. The broad features below $1500{\mathrm{cm}}^{-1}$ and at approximately $2220{\mathrm{cm}}^{-1}$ appear in the spectrum of the uncoated waveguide and when pumping at different wavelengths and are therefore identified as primarily due to Raman scattering in the ${\mathrm{SiO}}_2$ bottom cladding and the ${\mathrm{Si}}_3{\mathrm{N}}_4$ core of the waveguide. For comparison, bulk Raman scattering for ${\mathrm{SiO}}_2$ and amorphous ${\mathrm{Si}}_3{\mathrm{N}}_4$ below $1500{\mathrm{cm}}^{-1}$ can be found in [18,19] and [20], respectively. The broad feature centered around $2220{\mathrm{cm}}^{-1}$ is associated with Si-H resonances [21] within ${\mathrm{Si}}_3{\mathrm{N}}_4$. All other features in the ${\mathrm{N}}_2$-only spectrum are associated with the HCSFA2 polymer. These features appear unchanged in the spectrum during DMSO exposure except in the C-H stretch region just under $3000{\mathrm{cm}}^{-1}$, where DMSO and HCSFA2 have nearly coincidental spectral features. The arrows in the figure indicate the strongest spectral features associated with sorbed-phase DMSO.

 

Fig. 2. Raman spectra collected from an HCSFA2-coated 2.0 μm-wide waveguide both prior to and during exposure to DMSO. The spectrum during exposure to DMSO is shifted upward for clarity. The inset shows an enlargement of the range $650–725{\mathrm{cm}}^{-1}$ with no upward shift for the DMSO spectrum.

Download Full Size | PDF

To remove the signal not associated with the Raman scattering of the sorbed analyte species, we first normalized the spectra to account for optical power throughput differences. Next, we subtracted the broad, low-wavenumber features and the ${\mathrm{Si}}_3{\mathrm{N}}_4$ feature at $2220{\mathrm{cm}}^{-1}$ from the data using a peak-fitting procedure. Then, the before- and during-exposure spectra were subtracted from each other, yielding the differential Raman spectra shown in Fig. 3.

 

Fig. 3. Differential Raman spectra collected due to vapor-phase exposure to DMSO, MeS, and EA at the concentrations indicated. The spectrum above the graph for each analyte is the liquid-phase Raman spectrum from [22–24]. The concentration dependence for the boxed Raman line is shown to the right of each graph.

Download Full Size | PDF

4. DISCUSSION AND ANALYSIS

Liquid-phase Raman spectra from the literature are shown above each of our measured difference Raman spectrum in Fig. 3. As can be seen, there is excellent visual agreement between each of our observed sorbed-phase spectral lines and the known Raman lines for the liquid phase of the analyte. The central positions of the Raman lines identified in the difference spectrum for each analyte and the corresponding liquid-phase reference values are shown in Table 1. The differences in the measured versus literature values are presumptively due to differences in the sorbed- versus liquid-phase Raman signals. It is interesting to note that the measured numbers for DMSO, the analyte forming the strongest hydrogen bonding interaction with HCSFA2, are the most shifted from the literature values. It is known that the strength of the HB in sorbents can strongly affect the value of the polymer’s O-H stretch resonance [16,17]. Our results further suggest that individual sorbed-phase Raman lines of the most strongly bonded analytes may differ from those of the published liquid spectra. A full database for sorbed-phase Raman would therefore be required for spectral fingerprinting. The observation of multiple signature lines observed for each analyte studied in this work provides a suitable signal feature space for trace gas identification.

Table 1. Measured Raman Resonances (in ${\mathsf{\text{cm}}}^{-1}$) and Reference Values

View Table

To the right of each differential spectrum in Fig. 3, we show the concentration dependence for the boxed Raman line in the graph. These data are used to extrapolate the lowest concentration level we would expect to detect with this particular waveguide sample. For example, the smallest concentration we measured for DMSO was 135 ppb, but, even at this small concentration, the Raman line at $676{\mathrm{cm}}^{-1}$ is well above the noise level in the spectrum. Based on the measured concentration dependence, we have determined the one-sigma limit of detection (LOD) for DMSO as 7.6 ppb, for MeS as 360 ppb, and for EA as 600 ppm.

The Raman scattering efficiency for molecules near or inside a photonic waveguide is modified significantly compared to free space [12]. An accurate calculation of the scattering efficiency must account for both the modified density of the scattering modes as well as the collection efficiency into the waveguide modes. For a waveguide short enough to neglect loss and considering only the ${\mathrm{TE}}_{00}$ mode, we calculate the modified efficiency, which is defined as the ratio of forward-scattered Raman power to the pump power, to be [25]

Assuming equilibrated sorption, the analyte number density in the sorbent is given by $N_{\mathrm{HC}}=K{N}_{\mathrm{vapor}}$, where $K$ is the partition coefficient for the particular analyte under investigation, and $N_{\mathrm{vapor}}$ is the vapor number density above the waveguide. For DMSO at 500 ppb, $N_{\mathrm{vapor}}$ is $1.2\times {10}^{17}{\mathrm{m}}^{-3}$ and $K\sim 1\times {10}^8$ [14]. If we use $\sigma =3.3\times {10}^{-33}{\mathrm{m}}^2$ [26] and Comsol Multiphysics to calculate the field overlap integrals and the group index, we get an efficiency of $\eta =1.5\times {10}^{-12}$, which is consistent with our measured efficiency of $1.0\times {10}^{-12}$. The measured efficiency was determined using the counts/s at $676{\mathrm{cm}}^{-1}$, the known responses of the detector and spectrometer, and the measured input-to-spectrometer loss to form the ratio of Stokes power to pump power in the waveguide.

Our calculations show that, neglecting insertion losses and collection efficiencies, waveguide-based Raman scattering is enhanced compared with the traditional micro-Raman technique (${\eta }_{\mu \mathrm{R}}$) by

Similar to other hypersorbent polymers [27], we expect phosphonate esters to provide larger partition coefficients in HCFSA2 when compared to sulphoxides, such as DMSO, resulting in lower LODs [16]. Preparations for testing our current devices using phosphonate esters are underway. The tradeoff for lower LODs, however, is that an increased partition coefficient in a sorbent material is typically accompanied by a longer equilibration time for sorption. Measured sorption and desorption dynamics are shown in Fig. 4. These data were acquired by taking a 20 s spectrum of the strongest Raman line for each analyte and recording a spectrum every 1–2 min during sorption and desorption. The sorption/desorption times for EA are too rapid for this measurement technique to resolve. For the current waveguides with a 1–2 μm-thick polymer cladding, the sorption (desorption) times for MeS and DMSO were $1.8\pm 0.4\min$ ($3.1\pm 0.6\min$) and $8.8\pm 0.4\min$ ($14.5\pm 1.5\min$), respectively. We believe that the longer than expected measured desorption time in DMSO is the result of incomplete purging of DMSO vapor in our vapor generator and does not reflect the actual equilibration time. It is important to note that equilibration times depend on both the chemical species as well as the sorbent thickness. We are currently pursuing new approaches to provide much thinner, more uniform polymer coatings in order to reduce the signal kinetics. A thinner cladding allows faster diffusion of the analyte throughout the sorbent but will not degrade the strength of the collected Raman signal for films thicker than the evanescent penetration depth of the mode into the sorbent, approximately 100 nm. Thus, we expect order-of-magnitude faster responses with a thinner sorbent coating with no adverse impact on signal strength.

 

Fig. 4. Sorption and desorption times for MeS and DMSO. Similar graphs are not shown for EA because the sorption and desorption times were faster than the spectrum collection interval.

Download Full Size | PDF

5. CONCLUSION

In this work, we have shown that chemically functionalized, highly evanescent waveguides can be used to measure Raman spectra from sorbed trace gases with interaction lengths less than 1 cm. The Raman signal is dramatically enhanced by employing a hypersorbent coating as the top cladding of the waveguide and by the continuous generation and collection of Raman signal along the entire length of the waveguide. In this current work, we have used a sorbent polymer specifically designed to reversibly concentrate nerve agents and other HB basic molecules, but the concept can be extended for the use of different sorbent materials provided they are transparent to the pump laser and the Stokes-shifted Raman signal. For the future of this technology, we envision foundry-scale fabrication of functionalized waveguide arrays, each having a different coating to sorb different classes of vapors or gas. We expect that multiple Raman signal peaks per analyte will enable multi-component gases to be identified, similar to current solid-phase and liquid-phase Raman systems. Lastly, the type of sorbent used in this work is well suited to gas component separation techniques similar to those used in gas chromatographic columns. Studies are currently underway to investigate the feasibility of integrating gas chromatography techniques and functionalized waveguide structures.